Volume 125, Issue 21 2020JD033101
Research Article
Open Access

Quantifying Hail and Lightning Risk Factors Using Long-Term Observations Around Australia

Andrew J. Dowdy

Corresponding Author

Andrew J. Dowdy

Climate Research Section, Bureau of Meteorology, Melbourne, Victoria, Australia

Correspondence to:

A. J. Dowdy,

[email protected]

Contribution: Conceptualization, Methodology, Software, Validation, Formal analysis, ​Investigation, Resources, Data curation, Writing - original draft, Writing - review & editing, Visualization, Supervision, Project administration, Funding acquisition

Search for more papers by this author
Joshua Soderholm

Joshua Soderholm

Climate Research Section, Bureau of Meteorology, Melbourne, Victoria, Australia

Atmospheric Observations Research Group, The University of Queensland, Brisbane, Queensland, Australia

Contribution: Methodology, Software, Validation, Formal analysis, ​Investigation, Resources, Data curation, Writing - review & editing, Visualization

Search for more papers by this author
Jordan Brook

Jordan Brook

Atmospheric Observations Research Group, The University of Queensland, Brisbane, Queensland, Australia

Contribution: Methodology, Software, Validation, Formal analysis, ​Investigation, Resources, Data curation, Writing - review & editing

Search for more papers by this author
Andrew Brown

Andrew Brown

Climate Research Section, Bureau of Meteorology, Melbourne, Victoria, Australia

Contribution: Methodology, Software, Validation, Formal analysis, ​Investigation, Resources, Data curation, Writing - review & editing, Visualization

Search for more papers by this author
Hamish McGowan

Hamish McGowan

Atmospheric Observations Research Group, The University of Queensland, Brisbane, Queensland, Australia

Contribution: Validation, ​Investigation, Writing - review & editing, Supervision, Project administration

Search for more papers by this author
First published: 05 October 2020
Citations: 7

Abstract

There is a growing need to better understand and quantify risks associated with extreme weather, including severe thunderstorm-related hazards such as hail and lightning. Hail occurrence based on a long-term archive of radar observations is presented for the first time in many temperate and subtropical regions of Australia, together with lightning observations from a ground-based network of sensors. Mean monthly and hourly occurrence frequencies are examined for hail and lightning. Environmental conditions obtained from hourly reanalysis data indicate stronger wind shear on average for hail than lightning. The environmental conditions also indicate higher freezing levels on average for lightning than hail. These environmental differences provide plausible physical reasons for observed differences between hail and lightning climatology through the year. The study results are intended to help inform future planning and preparedness for thunderstorm-related risks, including for severe weather forecasting and climate risk applications.

Key Points

  • Occurrence frequencies of hail and lightning are presented based on observations for a wide range of regions around Australia
  • Environmental conditions are found to provide insight on spatiotemporal variations in hail and lightning activity
  • Environments associated with individual hazard types are recommended for use in improved preparedness for thunderstorm-related damages

Plain Language Summary

Natural hazards caused by thunderstorms, such as hail and lightning, can have severe impacts on society. Radar data are used to examine hail events at 10 Australian locations, and comparisons are made with lightning observations. Results are averaged over several years to calculate average annual values, including for the hourly and the monthly number of the hail and the lightning events. We find differences between hail and lightning events in the environmental conditions that they occur in. These environmental differences are found to help explain differences through the year in the average risk of occurrence of hail or lightning. The results are intended to enhance resilience in relation to thunderstorm-related risks and provide guidance for extreme weather modeling and climate adaptation purposes.

1 Introduction

Severe thunderstorms cause various meteorological hazards including extreme rainfall, damaging winds, tornadoes, hail, and lightning. These convection-related hazards can cause billions of dollars (United States Dollars: USD) annually in damages to property, infrastructure, and crops as well as thousands of deaths globally (Botzen et al., 2010; Brown et al., 2015; Changnon & Burroughs, 2003; Kunz et al., 2018; Púčik et al., 2019). Damages from individual hailstorms can exceed 1 billion USD, e.g., for Germany (Kunz et al., 2018), for the United States (Brown et al., 2015), and for Australia (Schuster et al., 2005), and cause fatalities when hailstones approach and exceed 5 cm (Calianese et al., 2002). Lightning can directly cause human deaths and is also the primary natural cause of wildfire ignition through the world, with lightning-ignited wildfires responsible for a large proportion of area burnt in regions such as Australia (Dowdy & Mills, 2012; Russell-Smith et al., 2007).

There is a need for improved preparedness for convection-related hazards including through greater understanding of their meteorology, mean occurrence frequencies, and environmental factors influencing their spatiotemporal variability. In Australia, previous studies have investigated the mean occurrence frequency of hail through analysis of radar observations near Brisbane and Sydney (Soderholm, McGowan, Richter, Walsh, Weckwerth, et al., 2017; Warren et al., 2020), satellite observations (Bedka et al., 2018; Cecil & Blankenship, 2012), as well as station observations and damage reports (Allen & Allen, 2016; Prein & Holland, 2018; Schuster et al., 2005, 2006). The mean occurrence frequency of lightning over the Australian region has been examined based on satellite observations (Bednarczyk & Sousounis, 2014; Dowdy & Kuleshov, 2014) and station observations (Bates et al., 2015). Additionally, environmental conditions based on reanalysis data have been used to identify risk factors associated with the occurrence of various thunderstorm-related hazards (Allen & Karoly, 2014; Bednarczyk & Sousounis, 2014; Dowdy, 2020; Niall & Walsh, 2005; Prein & Holland, 2018; Westermayer et al., 2017). However, important knowledge gaps remain around risks associated with thunderstorm-related hazards, including in relation to mean occurrence frequencies for different types of hazards and the factors influencing their variability.

This study aims to address several important gaps in knowledge, including through quantifying the mean occurrence frequencies of hail and lightning events for a range of different regions around Australia, as well as providing insight on factors influencing their spatiotemporal variability. The study compares the average characteristics of these two types of thunderstorm hazards, as well as the environmental conditions associated with their occurrence at a fine (hourly) temporal scale. Maximum Estimated Size of Hail (MESH) (Witt et al., 1998) derived from analysis of weather radar data is used to identify hail events, similar to approaches applied previously in Australia (Soderholm, McGowan, Richter, Walsh, Weckwerth, et al., 2017; Warren et al., 2020), the United States (Cintineo et al., 2012), Switzerland (Nisi et al., 2016), and Canada (Brimelow & Taylor, 2017). For comparison with the hail data, lightning observations from the World Wide Lightning Location Network (WWLLN) were obtained (Hutchins et al., 2013; Virts et al., 2013). Environmental data with hourly time steps from the Bureau of Meteorology Atmospheric high-resolution Regional Reanalysis for Australia (BARRA) reanalysis product produced by the Australian Bureau of Meteorology (Su et al., 2019) are used to identify environmental characteristics correlated with diurnal variations in hail and lightning.

2 Data and Methods

Radar data are used here from 10 locations around Australia with radar details listed in Table 1, including latitude, longitude, altitude, radar transmitter type, beamwidth, range resolution, scan time, and period of data used in this study. All times are in UTC throughout this study, noting that Local Time (not including Daylight Savings) is 10 hr ahead of UTC in all locations except Adelaide (9.5 hr ahead of UTC) and Perth (8 hr ahead of UTC).

Table 1. Details of the 10 Radars Including Radar Name (and Nearby City Referred to in This Study), Latitude, Longitude, Altitude, Transmitter Type, Angular Beamwidth, Range Resolution, Scan Time, and Period of Data Used for This Study
Radar name (city/region) Latitude, longitude Altitude of dish above mean sea level (m) Transmitter type Beamwidth Range resolution (m) Scan time (min) Period used for study
Mt. Kanigan (Gympie) −25.957, 152.577 375 S 1.9° Prior 1 December 2016: 500; after 2 December 2016: 250 Prior 1 December 2016: 10; after 2 December 2016: 6 2008–2018
Marburg (Brisbane) −27.606, 152.54 372 S 1.9° 1,000 10 1997–2018
Grafton (Grafton) −29.621, 152.963 40 S 1.9° 1,000 10 1998–2018
Namoi (Tamworth) −31.024, 150.192 699 S 1.9° 250 6 2010–2018
Serpentine (Perth) −32.392, 115.867 40 C 1.0° 500 (except 30 October 2014 to 1 December 2015 where it was 250) Prior 19 December 2015: 10; after 2 January 2015: 6 2010–2018
Wollongong (Sydney) −34.263, 150.875 460 S 1.9° Prior 25 September 2010: 1,000; after 5 February 2011: 500 Prior 25 September 2010: 10; after 5 February 2011: 6 1997–2018
Sellicks Hill (Adelaide) −35.33, 138.503 395 C 1.7° Prior 15 November 2014: 1,000; after 15 January 2015: 500 10 2005–2018
Captains Flat (Canberra) −35.661, 149.512 1,383 S 1.9° 500 Prior 2 March 2013: 10; after 18 May 2013: 6 2002–2018
Laverton (Melbourne) −37.855, 144.755 44 Prior 21 January 2007: C; after 25 August 2007: S 1.0° Prior 14 October 2004: 1,000; and prior 21 January 2007: 500; after 25 August 2007: 250 Prior 21 January 2007: 10; after 25 August 2007: 6 1998–2018
Mt. Koonya (Hobart) −43.113, 147.805 511 C 1.0° 250 6 2012–2018

The 10 radar sites were selected based on having a reasonably long period of available data, as needed for producing the climatological mean characteristics examined in this study. The 10 sites are in subtropical and extratropical locations through Australia, in regions around all of Australia's capital cities (apart from Darwin in the tropics), as shown in Figure 1. Tropical regions in the north of Australia are not examined here because MESH is not as well tested in those regions as compared to subtropical and temperate regions (Brimelow & Taylor, 2017; Cintineo et al., 2012; Nisi et al., 2016; Ortega, 2018). Although MESH estimates of hail climatology need to be treated carefully, such as noted by Murillo et al. (2020), MESH has been evaluated using a property insurance data set for the subtropical climate of Brisbane demonstrating that a MESH value of 32 mm was a skillful threshold for predicting losses from damaging hail cases (Warren et al., 2020). That threshold is comparable to the MESH threshold used by Cintineo et al. (2012) for indicating severe hail reports (>25 mm diameter) in the continental United States. It is also noted that the underlying integration in MESH of high reflectivity above the melting layer physically relates to hail, with changes to the depth of the warm layer (>0°C) in response to different climate zones expected to have limited impact on melting of severe hail stones (see, e.g., Fig. 2 of Ryzhkov et al., 2013).

Details are in the caption following the image
Map of the 10 radar locations, including showing their maximum 100-km range as used in this study (shaded white circular areas) and the radar region name in orange text (as listed in Table 1). Latitude and longitude are also shown, as well as state and territory capital city names in white text.

Prior to the retrieval of hail occurrence data from weather radar observations, a series of preliminary processing steps are applied. First, an absolute calibration of ground radar reflectivity is calculated using coincident measurements from the space-borne precipitation radars on the Tropical Rainfall Measuring Mission (TRMM) and Global Precipitation Measurement (GPM) Mission platforms (Louf et al., 2018; Warren et al., 2020). For each of the ground radars, their beam widths, elevation angles, and range resolutions are used to determine intersecting samples with satellite radar measurements. Common sample volumes that approximate intersecting regions are derived, and mean reflectivity measurements within the common volumes are calculated for each instrument, using the same set of requirements as applied in Warren et al. (2018, 2020). These requirements included a maximum delay between satellite and ground measurements of 300 s and a minimum of 10 satellite precipitation profiles within the ground radar domain. The reflectivity of the space-borne radar is transformed into the ground radar band (C or S) using the T-matrix calculations documented by Louf et al. (2018) to remove any band-dependent scattering biases. The calibration offset is then calculated as the difference between the mean of ground and space-borne radar samples. A more detailed description of the volume matching technique can be found in Warren et al. (2018).

Mean offset values between periods of stable radar operation (determined through examination of clutter stability and engineering logs) were applied to correct the ground radar reflectivity. Corrected volumetric reflectivity is then interpolated onto a Cartesian grid of 1 km in the horizontal dimension and 500 m in the vertical dimension (to 20 km altitude) using a Barnes weighting function (Barnes, 1964) for a constant radius of influence (2.5 km). Using the 3D grids of reflectivity and a coincident vertical temperature profile from ERA-Interim reanalysis (Dee et al., 2011), MESH is derived (Witt et al., 1998). Grid points closer than 20 km and beyond 100 km range from each radar site are removed due to ineffective sampling from the radar scanning pattern. Both a 30-mm MESH threshold and a 30-km2 cell area threshold (defined as contiguous pixels with ≥30-dBZ column-maximum reflectivity) are applied to MESH detections to remove nonsevere and null hailstorms and MESH detections resulting from nonmeteorological origins (Cintineo et al., 2012), with this study method intended to focus on larger hail associated with severe thunderstorm that could be more likely to cause damage (rather than smaller hail such as diameters less than about 25 mm).

Calibrating ground-based radars using space-borne radar data have seen ongoing investigation and application over the last decade (Crisologo & Heistermann, 2019; Schwaller & Morris, 2011; Warren et al., 2018, 2020) and are now used operationally in Australia and implemented in open source radar software (https://docs.wradlib.org/en/stable/). Fig. A1 from Warren et al. (2020) demonstrates the sensitivity of MESH to calibration aspects, with a detailed description of this calibration technique using space-borne GPM and TRMM observations applied to ground-based radar data sets provided in Warren et al. (2018).

C band radars have been successfully applied for detecting severe hailstorm occurrence as detailed in some recent studies (Kunz & Kugel, 2015; Lukach et al., 2017; Stržinar & Skok, 2018). Given that attenuation is more significant at C band than S band and that resonant scattering can have a greater effect at lower radar wavelengths (i.e., around 5 cm for C band and 10 cm for S band), there is potential for some variation in the relative magnitude of MESH retrievals for different radar types. However, the study method and aims are intended to help reduce the effect of differences between radars where feasible. For example, this includes relatively general aims focused on broad-scale features including the hours with maximum occurrence frequencies during the day, as well as the months with maximum occurrence frequencies during the year. Additionally, hailstorm events are considered in this study using a relatively broad-scale method based on the identification in a given hour of at least one event anywhere within a 100-km radius from the radar location.

The varying range resolution, beamwidth, and volume sample time between sites influence a radar detection sensitivity for short duration and/or small-sized hailstorms. Sampling geometry differences such as these are partly mitigated during gridding by using a radius of influence sufficiently large to sample the data at relatively coarse resolution (i.e., 2.5-km radius intervals), as well as noting other efforts to account for systematic differences between radars including calibration to space-borne radar data as described above and the broad-scale approach to defining hail events (i.e., based on at least one event identified anywhere in the 100-km radius region during a given hour rather than examining finer spatial or temporal details).

Periods of missing radar data are also accounted for when calculating mean values throughout this study. This is done individually for each radar, with the mean values calculated as the total number of hail events indicated by the radar data divided by the total period of available data. This process is also done for individual months of the year to calculate monthly mean values over the number of years available for a given radar.

For the purposes of this study, lightning events (2005–2018) are considered as having more than 150 lightning strokes recorded within 100 km of the radar location in a single hour (i.e., calculated individually for each of the 24 hr on any given day). This threshold was selected to provide a broadly similar number of lightning events to the number of hail events, while noting that the study aims do not require the number of hail and lightning events to be the same (e.g., with a focus on comparing the relative timing of peaks in the lightning and hail occurrence frequency distributions). Although changes in detection efficiency for the WWLLN network might potentially have some influence on the number of lightning events identified over time, the conclusions of the study were found to be robust to changes in the number of lightning events identified based on using either 100 or 200 lightning strokes in a given hour as the threshold for defining lightning events (i.e., below and above the threshold of 150 lightning strokes used in this study), with two different versions of the study results provided in Figures S1S6. Those alternative figures have varying numbers of lighting events being counted as expected based on varying the threshold used for defining those events but still show similar spatial and temporal features to the results presented for the threshold of 150 lightning strokes in a given hour (as discussed further for individual figures in the following results section).

Although long-term trend analysis is not a primary focus of this study, it is noted that a significant change is not able to be detected in hail occurrence for these 10 locations, including when considering changes in equipment over time, large interannual variability and the limited period of available data (Table 1). Similarly, for the WWLLN lightning data used here, there are small changes in equipment over time which influence detection efficiency, such that these data are not temporally homogenous to a sufficient degree as needed for trend analysis. However, the WWLLN data are suitable for climatological analyses, including with hourly time steps, as demonstrated in studies such as Virts et al. (2013). Additionally, as noted above for the hail events, results for the lightning events are interpreted throughout this study with consideration of the strengths and limitations of the available data, with conclusions presented based on the broad-scale climatological features including a focus on the hours with maximum occurrence frequencies during the day and months with maximum occurrence frequencies during the year.

Environmental conditions are examined based on the BARRA reanalysis data set (Su et al., 2019) with a spatial grid of about 12 km in both latitude and longitude as well as 1-hourly time steps used throughout this study. BARRA reanalysis has been used previously for the calculation of convective parameters such as those used in this study including cross-validation of such parameters from BARRA against observation-based data (Brown & Dowdy, 2019). It is based on the modeling approach used by the Bureau of Meteorology for operational weather forecasting in Australia, including similar data assimilation techniques as well as use of the UKMO Unified Model and parameterization schemes as detailed in Su et al. (2019). Additionally, to help demonstrate the robustness of results to different choices of reanalyses, the Supporting Information S1 includes a version of the results based on the ERA5 reanalysis (Hersbach et al., 2020) for comparison with BARRA reanalysis, as discussed in section 3.3.

Five different diagnostics are considered here, selected to be of relevance for operational severe weather forecasters in the Bureau of Meteorology and other agencies elsewhere in the world who routinely use such environmental diagnostics: convective available potential energy (CAPE), significant hail parameter (SHIP), bulk wind shear from 0 to 6 km (S06), freezing level height (FLH), and a product of CAPE and S06 (referred to here as CS6). CS6 is calculated following Allen et al. (2014) and Dowdy (2020) applications for examining Australian thunderstorms, based on earlier studies for the United States (Brooks et al., 2003), as shown by Equation 1:
urn:x-wiley:2169897X:media:jgrd56550:jgrd56550-math-0001(1)
CAPE is calculated here using a mixed-layer definition, whereby the starting parcel is defined by the average conditions between the surface and 100 hPa above ground level. This 100-hPa parcel layer for CAPE is intended to help allow for a range of storm types including those which may be supported by elevated inflow layers, such as can occur in coastal environments where the sea breeze may often be too stable, while noting that a broad range of CAPE definitions have been found to be useful in previous studies for various regions of the world (Allen et al., 2014; Brooks et al., 2003; Brown & Dowdy, 2019; Soderholm, McGowan, Richter, Walsh, Wedd, et al., 2017; Taszarek et al., 2018). The wrf-python software package was utilized for these calculations (Ladwig, 2017) with these standard formulations used for calculating the climatology values in this study. The definition of SHIP includes conditional requirements and follows the method from the U.S. Storm Prediction Center (SPC) from the U.S. National Oceanic and Atmospheric Administration (NOAA) calculated as follows:
urn:x-wiley:2169897X:media:jgrd56550:jgrd56550-math-0002(2)
where MUCAPE is the most unstable CAPE, MUq is the mixing ratio of the most unstable parcel (g·kg−1), LR is the temperature lapse rate between 700 and 500 hPa (°C·km−1), and T500 is air temperature at 500 hPa (°C). Further details can be found in NOAA documentation available online (https://www.spc.noaa.gov/exper/mesoanalysis/help/help_sigh.html), noting that SHIP values around 1 or higher are typically used for indicating environments associated with hail greater than about 2 inches (which is substantially larger than the hail sizes intended for this study based on a 30-mm MESH threshold as discussed above).

3 Results

3.1 Diurnal Characteristics

Figure 2 shows the diurnal variation in the occurrence frequency of hail and lightning events. Multiple events during a single hour at a given location are counted only once, with each hour representing the 60-min period following the start of the hour listed. The magnitudes vary with locations, with larger values typically occurring in the more eastern locations (particularly for Brisbane and Grafton) as compared to the more southern and western locations where relatively few events occur (e.g., Adelaide, Perth, and Hobart) broadly similar to previous studies of severe thunderstorm activity in these regions (e.g., Allen & Allen, 2016). Although there is considerable variation in magnitude with locations, the timing of the maxima and minima show less variation. Both the hail and lightning events most frequently occur from around 0400–0700 UTC (representing the mid-afternoon period in Local Time). There is some skewness of the hourly occurrence frequency distributions, with more events typically occurring after the time of peak occurrence frequency than before that time. Minimum values occur around 1700–2000 UTC for the lightning and hail events. These features are also seen when different thresholds are used for defining the lightning events, as shown in Figures S1 and S2.

Details are in the caption following the image
Diurnal variation in the mean occurrence frequency of hail (blue) and lightning (red) events for the (a–j) 10 radar regions. Annual mean number of events is shown for each hour. Dotted lines show a 3-point moving average to indicate the broader-scale climatology.

On average across all locations, 37% of hourly hail events also had a lightning event during the same hour, indicating a considerable degree of independence between these two different thunderstorm hazards types. Lower percentages than this were found for lightning events in earlier or later hours (i.e., weaker correlations for all time lags ≥1 hr in magnitude with respect to the hour of each hail event) with the strongest correlation for lag = 0 between lightning and hail occurrences. This indicates that there is not a large systematic difference in timing between individual hail and lightning events based on this analysis of hourly data.

3.2 Seasonal Variation

Figure 3 shows the variation through the year in the occurrence frequency of hail and lightning events. Multiple events recorded during a single day at a given location are counted only once. The results show considerable variation in magnitude between the different locations. There is relatively little variation between the different locations in the timing of the maxima and minima. The hail and lightning events occur more frequently during the warmer months of the year, broadly similar to a range of previous studies (Allen & Allen, 2016; Allen & Karoly, 2014; Bedka et al., 2018; Bednarczyk & Sousounis, 2014; Dowdy & Kuleshov, 2014; King & Kennedy, 2019; Schuster et al., 2005; Soderholm, McGowan, Richter, Walsh, Weckwerth, et al., 2017; Taszarek et al., 2017; Virts et al., 2013; Warren et al., 2020). The minimum values occur around the May to July period for the lightning and hail events.

Details are in the caption following the image
Monthly mean occurrence frequency of hail (blue) and lightning (red) events for the (a–j) 10 radar regions. Annual mean number of events is shown for each month. Dotted lines show a 3-point moving average to indicate the broader-scale climatology.

The results also indicate a somewhat earlier timing of hail events as compared to lightning events. The monthly cycle and timing differences are broadly similar at most locations, while noting relatively few lightning events for Melbourne prior to the summer solstice which accentuates this feature of the earlier timing of hail events as compared to lightning events. There is relatively little skewness in the shape of these monthly mean values, which are broadly symmetric about the month of maximum occurrence frequency. These features are also seen when different thresholds are used for defining the lightning events, as shown in Figures S3 and S4. The following section on environmental conditions also provides additional details on the variability of the hail and lightning climatological features.

3.3 Environmental Conditions

Environmental conditions are examined here based on five commonly used indices associated with thunderstorm occurrence as detailed in section 2: CAPE, SHIP, S06, CS6, and FLH. Figure 4 shows hourly and monthly mean values of these indices for the 100-km region around the radar location. The indices are calculated based on BARRA reanalysis data (Su et al., 2019) during the period 2005–2018. This period was selected based on the available period of lightning data. Data are used from all six locations with data available from 2005 to 2018 (Table 1): Grafton, Brisbane, Sydney, Adelaide, Canberra, and Melbourne. These six locations are all relatively similar in longitude (e.g., less than 1 hr different in Local Time for a given time in UTC).

Details are in the caption following the image
Diurnal and seasonal variation in five environmental parameters (a and f, CAPE; b and g, SHIP; c and h S06; d and i CS6; and e and j FLH) based on BARRA reanalysis data from 2005 to 2018. Average values are shown at the time of hail (blue) and lightning (red) events, as well as for all time steps (black), presented for each (a–e) hour and (f–j) month. These results are based on data from the six locations with observations available back to 2005 (Table 1). To help indicate the broader-scale features, the averaging includes ±1 time step for each value shown (i.e., ±1 hr for the hourly information and ±1 month for the monthly results). Results are only shown if at least 15 values contribute to a given average value, with a confidence range indicated by one standard error of the mean shown above and below the results (dotted lines).

Results are shown based on hourly time steps for hail events (blue) and lightning events (red), as well as for all hourly time steps from 2005 to 2015 (black). Results are shown only for cases where at least 15 individual values were available for calculating the mean values, to focus on results based on a reasonably large sample size. Although this choice of at least 15 values for the sample size is somewhat arbitrary, broadly similar results are found for moderate variations to this approach (e.g., for sample sizes in the range 10–20). All values are shown together with one standard error of the mean both above and below (dotted lines), as a measure based on the spread of the data and sample size (i.e., standard error of the mean calculated as the standard deviation divided by the square root of the number of events for each binning used for the values shown in Figure 4).

The CAPE, SHIP, and CS6 environmental measures have higher values for the thunderstorm event types (e.g., CAPE values on average around 500–1,000 J·kg−1 for the hail events and the lightning events) than for all time steps (e.g., close to 0 J·kg−1) from Figure 4. There is also some indication that lightning events have larger average values of CAPE, SHIP, and CS6 later in the day and into the evening as compared to the case for hail events which tend to have larger values earlier in the day for those three environmental measures. When considering the case for all time steps, these environmental measures show maxima around 0400–1000 UTC (i.e., during the afternoon and early evening in Local Time), as well as around the warmer months of the year. However, those maxima are small in magnitude compared to the case for the hail and lightning events which are about 10 times larger in all cases (Figures 4a, 4b, 4d, 4f, 4g, and 4i).

The SHIP and CS6 values are higher in general for the hail events than the lightning events, in contrast to the case for CAPE. Although SHIP and CS6 are based on CAPE, they also include wind shear in their formulations (section 2), indicating a greater dependence on wind shear for hail than for lightning events. This is consistent with results based on examining the bulk wind S06 showing larger mean values for the hail events than the lightning events (by about 4 m·s−1 on average from Figures 4c and 4h), noting some similarities to studies including based on modeling and observations data for hail events (Dennis & Kumjian, 2017; Johnson & Sugden, 2014). It is noted that the S06 results for all times show larger values during the cooler months than the warmer month of the year.

The results also show that SHIP values may be somewhat better for predicting hail than lightning, given that the values are generally higher for hail events than lightning events. This difference in mean SHIP values is largest during the afternoon period in Local Time (i.e., corresponding to about 0–6 UTC) with hail events having values of about 0.6–0.7 and lightning events having values of about 0.4–0.5 from Figure 4b. In contrast, for the period later in the afternoon and into the evening, hail events tend to have relatively large values of wind shear (Figure 4c) and CAPE values not as high (Figure 4a) as for lightning events. This has implications for the prediction of thunderstorms in relatively low-CAPE environments, including showing that these tend to occur with strong wind shear in the late afternoon and first half of the night and are more often associated with the occurrence of the hail events than the lightning events.

FLH values are higher for the thunderstorm event types as compared to all time steps, particularly prior to the austral summer with values about 0.3–0.6 km higher on average for hail and lightning events than for all time steps (Figure 4j). The lightning events tend to have higher values of FLH as compared to the hail events, particularly for later times in the day (after about 0300 UTC corresponding to the mid-afternoon and evening in Local Time) with differences of the order of 0.3–0.6 km on average, as well as from about December to June (i.e., during the period from the austral summer until winter) with smaller differences of the order of 0.1–0.2 km on average. When considering the case for all times, lower values of FLH generally occur prior to the austral summer solstice.

As shown here for the results based on all times (i.e., black lines in Figure 4), stronger wind shear and lower FLH occur more often prior to the austral summer solstice. This provides one plausible physical reason for why the hail events occur earlier with respect to the warm-season peak than the lightning events (Figure 3), given the stronger wind shear and lower FLH associated with the hail events than the lightning events (blue and red lines in Figure 4). It may also provide some insight on why Melbourne has a relatively low frequency of lightning occurrence as compared to hail prior to the summer solstice (Figure 3i). For example, this could occur if thunderstorm production in that region was strongly influenced by wind shear and FLH as well as not as strongly influenced by instability-related aspects as compared to the case in other regions (noting scope for potential future research into region-specific features such as this). A lower FLH could help reduce the chance of hail melting, with previous research reporting a change in hailstone size distributions for increased melting level height (Dessens et al., 2015), while also noting that the depth of the warm layer (>0°C) may not be a major factor for melting of larger hailstones as compared to other factors such as hail growth processes at higher altitudes (Allen et al., 2020; Mahoney et al., 2012; Ryzhkov et al., 2013).

This type of systematic long-term analysis of convective parameters derived from fine-resolution (e.g., hourly) reanalysis data has not previously been presented for Australia including for hail and lightning events over multiple years. Examples such as the FLH, S06, CAPE, SHIP, and CS6 characteristics discussed above from Figure 4 indicate features that could be helpful as part of guidance information around predictions of hail occurrence or lightning occurrence (i.e., for operational severe weather forecasters in the Bureau of Meteorology who routinely such environmental diagnostics including those examined here). As noted in section 2, the BARRA reanalysis used in this study is based on the same modeling approach used by the Bureau of Meteorology for operational weather forecasting in Australia. Additionally, broadly similar results to those described above for Figure 4 were found when a different reanalysis data set was used (i.e., for the ERA5 reanalysis data set in Figure S7).

4 Discussion and Conclusions

This study investigated various characteristics of hailstorms based on radar data and lightning observations. Mean hourly and monthly occurrence frequencies were presented at 10 locations through subtropical and extratropical regions of Australia. Hail and lightning events most frequently occurred during the mid-afternoon and in the warm period of the year, similar to previous studies (Allen & Karoly, 2014; Bedka et al., 2018; Bednarczyk & Sousounis, 2014; Dowdy & Kuleshov, 2014; King & Kennedy, 2019; Schuster et al., 2005; Soderholm, McGowan, Richter, Walsh, Weckwerth, et al., 2017; Taszarek et al., 2017; Virts et al., 2013; Warren et al., 2020), but the study results also show hail events occurring earlier with respect to the austral summer than lightning events.

Environmental conditions associated with thunderstorm activity were examined based on hourly reanalysis data. The diurnal and seasonal variations in both hail and lightning activity were found to be broadly similar to the diurnal and seasonal variations in CAPE as well as measures based on CAPE (including the SHIP and CS6 measures examined here). Hail events were found to have larger values for wind shear-related measures (S06, SHIP, and CS6) than lightning events, while lightning events tend to have larger values of FLH than hail events.

On average for all times, it was found that stronger wind shear and lower FLH occur more often prior to the warm-season peak in hail and lightning occurrence. These environmental conditions provide a plausible reason for why the hail events occur earlier with respect to the warm-season peak than the lightning events, given the stronger wind shear and lower FLH for the hail events than the lightning events on average.

During the period later in the afternoon and in the first half of the night, hail events were found to occur with relatively low CAPE values as compared to the case for the lightning events. This has implications for the prediction of thunderstorms in relatively low-CAPE environments, including showing that these events tend to occur with strong wind shear during the period in the late afternoon and first half of the night and are more often associated with the occurrence of the hail events than the lightning events.

The systematic comparison of environmental conditions associated with hail and lightning activity presented here provides information relevant for severe weather forecasting as well as climate modeling applications, including if considering a specific type of hazard such as hail or lightning as distinct from thunderstorm activity more generically. Additionally, future research could investigate large-scale modes of variability such as the El Niño-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD), and Southern Annular Mode (SAM) in relation to their influence on these environmental conditions and examine the potential for long-range (e.g., multiweek to seasonal) predictions of specific hazards such as hail or lightning (Allen & Karoly, 2014; Dowdy, 2016).

As longer data sets become available over time (i.e., due to more years of observations), there will be a greater ability to examine the influence of these large-scale modes of variability on thunderstorm-related hazards, such as for El Niño conditions as compared to La Niña conditions (i.e., for individual phases of ENSO). However, examining subsets of the data in that way is not done in this study, including given that the sample size is currently not very large for enabling such climatological analysis to be done with a reasonable degree of confidence. Similarly, larger sample sizes would also be beneficial for further examinations of spatiotemporal variations in thunderstorm-related hazards for individual regions or seasons (e.g., diurnal variation for summer as compared to winter in thunderstorm-related hazards and associated environmental conditions in specific regions). Future research could also examine the potential influence of aerosols on lightning and hail occurrences, including smoke from wildfires as a source of cloud condensation nuclei that might influence the microphysics and other thermodynamic processes for severe thunderstorms and associated hazards. In addition to aerosols, there is a wide range of other factors that can potentially influence thunderstorm occurrence. The five environmental conditions examined in this study are intended as a relatively concise set of diagnostics based on those commonly used including in previous studies, as well as in operational severe weather forecasting purposes (e.g., for the Australian Bureau of Meteorology and SPC/NOAA), while noting that future work could examine a broader set of diagnostics than those considered or this study.

Modeling studies of the future climate indicate that severe convective storms could potentially become more frequent in many of the heavily populated regions of Australia, including around cities such as Brisbane, Sydney, and Melbourne (Allen et al., 2014; Niall & Walsh, 2005), while noting considerable uncertainties around the potential influence of climate change on thunderstorms and associated hazards. An increase in the frequency of thermodynamic conditions favorable to deep and moist convection is consistent with results from a range of climate change studies, including as a response to increased lower tropospheric temperatures and atmospheric moisture capacity (noting the Clausius-Clapeyron relation in a warming climate) which can result in larger CAPE in some cases (Agard & Emanuel, 2017; Allen et al., 2020; Prein & Holland, 2018; Price & Rind, 1994; Romps, 2016).

An increase in tropospheric temperature due to climate change (IPCC, 2018) might increase FLH and thereby influence hail and lightning in somewhat different ways (e.g., given their different relationships to FLH from Figure 4, particularly in the afternoon and night as well as in months after the warm-season peak). Increased FLH could potentially result in a change in hailstone size distributions, noting an increase in mean kinetic energy of point hailfall for increased FLH demonstrated in previous studies (Dessens et al., 2015). Improved modeling of how climate change might influence environmental conditions such as those examined in this study (from Figure 4) could have benefits for preparedness and adaptation in relation to specific types of thunderstorm-related hazards in a warmer world (e.g., potential changes in hail activity that might be different to changes in lightning activity or other hazards such as extreme convective rainfall or wind events).

Our findings provide new insight on thunderstorm-related hazards in subtropical and extratropical regions where a large amount of Australia's population and infrastructure is located. Quantifying the spatiotemporal variations in the probability of occurrence for hail and lightning events, together with their associated environmental conditions, will help improve severe weather prediction services. This includes short-range forecasting applications in operational agencies (e.g., Bureau of Meteorology in Australia) as well as longer term planning around climate risk and disaster risk reduction. Improved preparedness for thunderstorm-related risks will help reduce their impacts on society, with considerable benefits for a range of sectors including in relation to emergency management, health, finance, and energy.

Acknowledgments

The authors declare no real or perceived conflict of interest with respect to the results of this paper. This work was supported by the ESCC Hub of the Australian Government's National Environmental Science Program (NESP). Comments on earlier drafts by Alain Protat and Surendra Rauniyar from Bureau of Meteorology are gratefully acknowledged.

    Data Availability Statement

    Data sets for this research are available as archived by the Australian Government's Bureau of Meteorology (Telephone +61396694000; Mail address 700 Collins St, Docklands, 461 3001, Victoria, Australia) including for the BARRA reanalysis data (with specific details on access to BARRA data also available here: http://www.bom.gov.au/research/projects/reanalysis/) as well as for hourly occurrences of the two thunderstorm event types for the ten study regions (with processed data also available from the Zenodo archive with reference doi:10.5281/zenodo.3977794). Data are available on request from Australian Bureau of Meteorology